14 typedef long long BLASLONG;
15 typedef unsigned long long BLASULONG;
17 typedef long BLASLONG;
18 typedef unsigned long BLASULONG;
22 typedef BLASLONG blasint;
24 #define blasabs(x) llabs(x)
26 #define blasabs(x) labs(x)
30 #define blasabs(x) abs(x)
33 typedef blasint integer;
35 typedef unsigned int uinteger;
36 typedef char *address;
37 typedef short int shortint;
39 typedef double doublereal;
40 typedef struct { real r, i; } complex;
41 typedef struct { doublereal r, i; } doublecomplex;
43 static inline _Fcomplex Cf(complex *z) {_Fcomplex zz={z->r , z->i}; return zz;}
44 static inline _Dcomplex Cd(doublecomplex *z) {_Dcomplex zz={z->r , z->i};return zz;}
45 static inline _Fcomplex * _pCf(complex *z) {return (_Fcomplex*)z;}
46 static inline _Dcomplex * _pCd(doublecomplex *z) {return (_Dcomplex*)z;}
48 static inline _Complex float Cf(complex *z) {return z->r + z->i*_Complex_I;}
49 static inline _Complex double Cd(doublecomplex *z) {return z->r + z->i*_Complex_I;}
50 static inline _Complex float * _pCf(complex *z) {return (_Complex float*)z;}
51 static inline _Complex double * _pCd(doublecomplex *z) {return (_Complex double*)z;}
53 #define pCf(z) (*_pCf(z))
54 #define pCd(z) (*_pCd(z))
56 typedef short int shortlogical;
57 typedef char logical1;
58 typedef char integer1;
63 /* Extern is for use with -E */
74 /*external read, write*/
83 /*internal read, write*/
113 /*rewind, backspace, endfile*/
125 ftnint *inex; /*parameters in standard's order*/
151 union Multitype { /* for multiple entry points */
162 typedef union Multitype Multitype;
164 struct Vardesc { /* for Namelist */
170 typedef struct Vardesc Vardesc;
177 typedef struct Namelist Namelist;
179 #define abs(x) ((x) >= 0 ? (x) : -(x))
180 #define dabs(x) (fabs(x))
181 #define f2cmin(a,b) ((a) <= (b) ? (a) : (b))
182 #define f2cmax(a,b) ((a) >= (b) ? (a) : (b))
183 #define dmin(a,b) (f2cmin(a,b))
184 #define dmax(a,b) (f2cmax(a,b))
185 #define bit_test(a,b) ((a) >> (b) & 1)
186 #define bit_clear(a,b) ((a) & ~((uinteger)1 << (b)))
187 #define bit_set(a,b) ((a) | ((uinteger)1 << (b)))
189 #define abort_() { sig_die("Fortran abort routine called", 1); }
190 #define c_abs(z) (cabsf(Cf(z)))
191 #define c_cos(R,Z) { pCf(R)=ccos(Cf(Z)); }
193 #define c_div(c, a, b) {Cf(c)._Val[0] = (Cf(a)._Val[0]/Cf(b)._Val[0]); Cf(c)._Val[1]=(Cf(a)._Val[1]/Cf(b)._Val[1]);}
194 #define z_div(c, a, b) {Cd(c)._Val[0] = (Cd(a)._Val[0]/Cd(b)._Val[0]); Cd(c)._Val[1]=(Cd(a)._Val[1]/df(b)._Val[1]);}
196 #define c_div(c, a, b) {pCf(c) = Cf(a)/Cf(b);}
197 #define z_div(c, a, b) {pCd(c) = Cd(a)/Cd(b);}
199 #define c_exp(R, Z) {pCf(R) = cexpf(Cf(Z));}
200 #define c_log(R, Z) {pCf(R) = clogf(Cf(Z));}
201 #define c_sin(R, Z) {pCf(R) = csinf(Cf(Z));}
202 //#define c_sqrt(R, Z) {*(R) = csqrtf(Cf(Z));}
203 #define c_sqrt(R, Z) {pCf(R) = csqrtf(Cf(Z));}
204 #define d_abs(x) (fabs(*(x)))
205 #define d_acos(x) (acos(*(x)))
206 #define d_asin(x) (asin(*(x)))
207 #define d_atan(x) (atan(*(x)))
208 #define d_atn2(x, y) (atan2(*(x),*(y)))
209 #define d_cnjg(R, Z) { pCd(R) = conj(Cd(Z)); }
210 #define r_cnjg(R, Z) { pCf(R) = conjf(Cf(Z)); }
211 #define d_cos(x) (cos(*(x)))
212 #define d_cosh(x) (cosh(*(x)))
213 #define d_dim(__a, __b) ( *(__a) > *(__b) ? *(__a) - *(__b) : 0.0 )
214 #define d_exp(x) (exp(*(x)))
215 #define d_imag(z) (cimag(Cd(z)))
216 #define r_imag(z) (cimagf(Cf(z)))
217 #define d_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
218 #define r_int(__x) (*(__x)>0 ? floor(*(__x)) : -floor(- *(__x)))
219 #define d_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
220 #define r_lg10(x) ( 0.43429448190325182765 * log(*(x)) )
221 #define d_log(x) (log(*(x)))
222 #define d_mod(x, y) (fmod(*(x), *(y)))
223 #define u_nint(__x) ((__x)>=0 ? floor((__x) + .5) : -floor(.5 - (__x)))
224 #define d_nint(x) u_nint(*(x))
225 #define u_sign(__a,__b) ((__b) >= 0 ? ((__a) >= 0 ? (__a) : -(__a)) : -((__a) >= 0 ? (__a) : -(__a)))
226 #define d_sign(a,b) u_sign(*(a),*(b))
227 #define r_sign(a,b) u_sign(*(a),*(b))
228 #define d_sin(x) (sin(*(x)))
229 #define d_sinh(x) (sinh(*(x)))
230 #define d_sqrt(x) (sqrt(*(x)))
231 #define d_tan(x) (tan(*(x)))
232 #define d_tanh(x) (tanh(*(x)))
233 #define i_abs(x) abs(*(x))
234 #define i_dnnt(x) ((integer)u_nint(*(x)))
235 #define i_len(s, n) (n)
236 #define i_nint(x) ((integer)u_nint(*(x)))
237 #define i_sign(a,b) ((integer)u_sign((integer)*(a),(integer)*(b)))
238 #define pow_dd(ap, bp) ( pow(*(ap), *(bp)))
239 #define pow_si(B,E) spow_ui(*(B),*(E))
240 #define pow_ri(B,E) spow_ui(*(B),*(E))
241 #define pow_di(B,E) dpow_ui(*(B),*(E))
242 #define pow_zi(p, a, b) {pCd(p) = zpow_ui(Cd(a), *(b));}
243 #define pow_ci(p, a, b) {pCf(p) = cpow_ui(Cf(a), *(b));}
244 #define pow_zz(R,A,B) {pCd(R) = cpow(Cd(A),*(B));}
245 #define s_cat(lpp, rpp, rnp, np, llp) { ftnlen i, nc, ll; char *f__rp, *lp; ll = (llp); lp = (lpp); for(i=0; i < (int)*(np); ++i) { nc = ll; if((rnp)[i] < nc) nc = (rnp)[i]; ll -= nc; f__rp = (rpp)[i]; while(--nc >= 0) *lp++ = *(f__rp)++; } while(--ll >= 0) *lp++ = ' '; }
246 #define s_cmp(a,b,c,d) ((integer)strncmp((a),(b),f2cmin((c),(d))))
247 #define s_copy(A,B,C,D) { int __i,__m; for (__i=0, __m=f2cmin((C),(D)); __i<__m && (B)[__i] != 0; ++__i) (A)[__i] = (B)[__i]; }
248 #define sig_die(s, kill) { exit(1); }
249 #define s_stop(s, n) {exit(0);}
250 static char junk[] = "\n@(#)LIBF77 VERSION 19990503\n";
251 #define z_abs(z) (cabs(Cd(z)))
252 #define z_exp(R, Z) {pCd(R) = cexp(Cd(Z));}
253 #define z_sqrt(R, Z) {pCd(R) = csqrt(Cd(Z));}
254 #define myexit_() break;
255 #define mycycle() continue;
256 #define myceiling(w) {ceil(w)}
257 #define myhuge(w) {HUGE_VAL}
258 //#define mymaxloc_(w,s,e,n) {if (sizeof(*(w)) == sizeof(double)) dmaxloc_((w),*(s),*(e),n); else dmaxloc_((w),*(s),*(e),n);}
259 #define mymaxloc(w,s,e,n) {dmaxloc_(w,*(s),*(e),n)}
261 /* procedure parameter types for -A and -C++ */
263 #define F2C_proc_par_types 1
265 typedef logical (*L_fp)(...);
267 typedef logical (*L_fp)();
270 static float spow_ui(float x, integer n) {
271 float pow=1.0; unsigned long int u;
273 if(n < 0) n = -n, x = 1/x;
282 static double dpow_ui(double x, integer n) {
283 double pow=1.0; unsigned long int u;
285 if(n < 0) n = -n, x = 1/x;
295 static _Fcomplex cpow_ui(complex x, integer n) {
296 complex pow={1.0,0.0}; unsigned long int u;
298 if(n < 0) n = -n, x.r = 1/x.r, x.i=1/x.i;
300 if(u & 01) pow.r *= x.r, pow.i *= x.i;
301 if(u >>= 1) x.r *= x.r, x.i *= x.i;
305 _Fcomplex p={pow.r, pow.i};
309 static _Complex float cpow_ui(_Complex float x, integer n) {
310 _Complex float pow=1.0; unsigned long int u;
312 if(n < 0) n = -n, x = 1/x;
323 static _Dcomplex zpow_ui(_Dcomplex x, integer n) {
324 _Dcomplex pow={1.0,0.0}; unsigned long int u;
326 if(n < 0) n = -n, x._Val[0] = 1/x._Val[0], x._Val[1] =1/x._Val[1];
328 if(u & 01) pow._Val[0] *= x._Val[0], pow._Val[1] *= x._Val[1];
329 if(u >>= 1) x._Val[0] *= x._Val[0], x._Val[1] *= x._Val[1];
333 _Dcomplex p = {pow._Val[0], pow._Val[1]};
337 static _Complex double zpow_ui(_Complex double x, integer n) {
338 _Complex double pow=1.0; unsigned long int u;
340 if(n < 0) n = -n, x = 1/x;
350 static integer pow_ii(integer x, integer n) {
351 integer pow; unsigned long int u;
353 if (n == 0 || x == 1) pow = 1;
354 else if (x != -1) pow = x == 0 ? 1/x : 0;
357 if ((n > 0) || !(n == 0 || x == 1 || x != -1)) {
367 static integer dmaxloc_(double *w, integer s, integer e, integer *n)
369 double m; integer i, mi;
370 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
371 if (w[i-1]>m) mi=i ,m=w[i-1];
374 static integer smaxloc_(float *w, integer s, integer e, integer *n)
376 float m; integer i, mi;
377 for(m=w[s-1], mi=s, i=s+1; i<=e; i++)
378 if (w[i-1]>m) mi=i ,m=w[i-1];
381 static inline void cdotc_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
382 integer n = *n_, incx = *incx_, incy = *incy_, i;
384 _Fcomplex zdotc = {0.0, 0.0};
385 if (incx == 1 && incy == 1) {
386 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
387 zdotc._Val[0] += conjf(Cf(&x[i]))._Val[0] * Cf(&y[i])._Val[0];
388 zdotc._Val[1] += conjf(Cf(&x[i]))._Val[1] * Cf(&y[i])._Val[1];
391 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
392 zdotc._Val[0] += conjf(Cf(&x[i*incx]))._Val[0] * Cf(&y[i*incy])._Val[0];
393 zdotc._Val[1] += conjf(Cf(&x[i*incx]))._Val[1] * Cf(&y[i*incy])._Val[1];
399 _Complex float zdotc = 0.0;
400 if (incx == 1 && incy == 1) {
401 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
402 zdotc += conjf(Cf(&x[i])) * Cf(&y[i]);
405 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
406 zdotc += conjf(Cf(&x[i*incx])) * Cf(&y[i*incy]);
412 static inline void zdotc_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
413 integer n = *n_, incx = *incx_, incy = *incy_, i;
415 _Dcomplex zdotc = {0.0, 0.0};
416 if (incx == 1 && incy == 1) {
417 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
418 zdotc._Val[0] += conj(Cd(&x[i]))._Val[0] * Cd(&y[i])._Val[0];
419 zdotc._Val[1] += conj(Cd(&x[i]))._Val[1] * Cd(&y[i])._Val[1];
422 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
423 zdotc._Val[0] += conj(Cd(&x[i*incx]))._Val[0] * Cd(&y[i*incy])._Val[0];
424 zdotc._Val[1] += conj(Cd(&x[i*incx]))._Val[1] * Cd(&y[i*incy])._Val[1];
430 _Complex double zdotc = 0.0;
431 if (incx == 1 && incy == 1) {
432 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
433 zdotc += conj(Cd(&x[i])) * Cd(&y[i]);
436 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
437 zdotc += conj(Cd(&x[i*incx])) * Cd(&y[i*incy]);
443 static inline void cdotu_(complex *z, integer *n_, complex *x, integer *incx_, complex *y, integer *incy_) {
444 integer n = *n_, incx = *incx_, incy = *incy_, i;
446 _Fcomplex zdotc = {0.0, 0.0};
447 if (incx == 1 && incy == 1) {
448 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
449 zdotc._Val[0] += Cf(&x[i])._Val[0] * Cf(&y[i])._Val[0];
450 zdotc._Val[1] += Cf(&x[i])._Val[1] * Cf(&y[i])._Val[1];
453 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
454 zdotc._Val[0] += Cf(&x[i*incx])._Val[0] * Cf(&y[i*incy])._Val[0];
455 zdotc._Val[1] += Cf(&x[i*incx])._Val[1] * Cf(&y[i*incy])._Val[1];
461 _Complex float zdotc = 0.0;
462 if (incx == 1 && incy == 1) {
463 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
464 zdotc += Cf(&x[i]) * Cf(&y[i]);
467 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
468 zdotc += Cf(&x[i*incx]) * Cf(&y[i*incy]);
474 static inline void zdotu_(doublecomplex *z, integer *n_, doublecomplex *x, integer *incx_, doublecomplex *y, integer *incy_) {
475 integer n = *n_, incx = *incx_, incy = *incy_, i;
477 _Dcomplex zdotc = {0.0, 0.0};
478 if (incx == 1 && incy == 1) {
479 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
480 zdotc._Val[0] += Cd(&x[i])._Val[0] * Cd(&y[i])._Val[0];
481 zdotc._Val[1] += Cd(&x[i])._Val[1] * Cd(&y[i])._Val[1];
484 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
485 zdotc._Val[0] += Cd(&x[i*incx])._Val[0] * Cd(&y[i*incy])._Val[0];
486 zdotc._Val[1] += Cd(&x[i*incx])._Val[1] * Cd(&y[i*incy])._Val[1];
492 _Complex double zdotc = 0.0;
493 if (incx == 1 && incy == 1) {
494 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
495 zdotc += Cd(&x[i]) * Cd(&y[i]);
498 for (i=0;i<n;i++) { /* zdotc = zdotc + dconjg(x(i))* y(i) */
499 zdotc += Cd(&x[i*incx]) * Cd(&y[i*incy]);
505 /* -- translated by f2c (version 20000121).
506 You must link the resulting object file with the libraries:
507 -lf2c -lm (in that order)
513 /* Table of constant values */
515 static integer c__1 = 1;
516 static integer c_n1 = -1;
517 static doublereal c_b12 = 1.;
518 static doublereal c_b13 = 0.;
519 static doublereal c_b21 = -1.;
521 /* > \brief \b DSYTRF_AA_2STAGE */
523 /* =========== DOCUMENTATION =========== */
525 /* Online html documentation available at */
526 /* http://www.netlib.org/lapack/explore-html/ */
529 /* > Download DSYTRF_AA_2STAGE + dependencies */
530 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.tgz?format=tgz&filename=/lapack/lapack_routine/dsytrf_
533 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.zip?format=zip&filename=/lapack/lapack_routine/dsytrf_
536 /* > <a href="http://www.netlib.org/cgi-bin/netlibfiles.txt?format=txt&filename=/lapack/lapack_routine/dsytrf_
544 /* SUBROUTINE DSYTRF_AA_2STAGE( UPLO, N, A, LDA, TB, LTB, IPIV, */
545 /* IPIV2, WORK, LWORK, INFO ) */
548 /* INTEGER N, LDA, LTB, LWORK, INFO */
549 /* INTEGER IPIV( * ), IPIV2( * ) */
550 /* DOUBLE PRECISION A( LDA, * ), TB( * ), WORK( * ) */
552 /* > \par Purpose: */
557 /* > DSYTRF_AA_2STAGE computes the factorization of a real symmetric matrix A */
558 /* > using the Aasen's algorithm. The form of the factorization is */
560 /* > A = U**T*T*U or A = L*T*L**T */
562 /* > where U (or L) is a product of permutation and unit upper (lower) */
563 /* > triangular matrices, and T is a symmetric band matrix with the */
564 /* > bandwidth of NB (NB is internally selected and stored in TB( 1 ), and T is */
565 /* > LU factorized with partial pivoting). */
567 /* > This is the blocked version of the algorithm, calling Level 3 BLAS. */
573 /* > \param[in] UPLO */
575 /* > UPLO is CHARACTER*1 */
576 /* > = 'U': Upper triangle of A is stored; */
577 /* > = 'L': Lower triangle of A is stored. */
583 /* > The order of the matrix A. N >= 0. */
586 /* > \param[in,out] A */
588 /* > A is DOUBLE PRECISION array, dimension (LDA,N) */
589 /* > On entry, the symmetric matrix A. If UPLO = 'U', the leading */
590 /* > N-by-N upper triangular part of A contains the upper */
591 /* > triangular part of the matrix A, and the strictly lower */
592 /* > triangular part of A is not referenced. If UPLO = 'L', the */
593 /* > leading N-by-N lower triangular part of A contains the lower */
594 /* > triangular part of the matrix A, and the strictly upper */
595 /* > triangular part of A is not referenced. */
597 /* > On exit, L is stored below (or above) the subdiaonal blocks, */
598 /* > when UPLO is 'L' (or 'U'). */
601 /* > \param[in] LDA */
603 /* > LDA is INTEGER */
604 /* > The leading dimension of the array A. LDA >= f2cmax(1,N). */
607 /* > \param[out] TB */
609 /* > TB is DOUBLE PRECISION array, dimension (LTB) */
610 /* > On exit, details of the LU factorization of the band matrix. */
613 /* > \param[in] LTB */
615 /* > LTB is INTEGER */
616 /* > The size of the array TB. LTB >= 4*N, internally */
617 /* > used to select NB such that LTB >= (3*NB+1)*N. */
619 /* > If LTB = -1, then a workspace query is assumed; the */
620 /* > routine only calculates the optimal size of LTB, */
621 /* > returns this value as the first entry of TB, and */
622 /* > no error message related to LTB is issued by XERBLA. */
625 /* > \param[out] IPIV */
627 /* > IPIV is INTEGER array, dimension (N) */
628 /* > On exit, it contains the details of the interchanges, i.e., */
629 /* > the row and column k of A were interchanged with the */
630 /* > row and column IPIV(k). */
633 /* > \param[out] IPIV2 */
635 /* > IPIV2 is INTEGER array, dimension (N) */
636 /* > On exit, it contains the details of the interchanges, i.e., */
637 /* > the row and column k of T were interchanged with the */
638 /* > row and column IPIV2(k). */
641 /* > \param[out] WORK */
643 /* > WORK is DOUBLE PRECISION workspace of size LWORK */
646 /* > \param[in] LWORK */
648 /* > LWORK is INTEGER */
649 /* > The size of WORK. LWORK >= N, internally used to select NB */
650 /* > such that LWORK >= N*NB. */
652 /* > If LWORK = -1, then a workspace query is assumed; the */
653 /* > routine only calculates the optimal size of the WORK array, */
654 /* > returns this value as the first entry of the WORK array, and */
655 /* > no error message related to LWORK is issued by XERBLA. */
658 /* > \param[out] INFO */
660 /* > INFO is INTEGER */
661 /* > = 0: successful exit */
662 /* > < 0: if INFO = -i, the i-th argument had an illegal value. */
663 /* > > 0: if INFO = i, band LU factorization failed on i-th column */
669 /* > \author Univ. of Tennessee */
670 /* > \author Univ. of California Berkeley */
671 /* > \author Univ. of Colorado Denver */
672 /* > \author NAG Ltd. */
674 /* > \date November 2017 */
676 /* > \ingroup doubleSYcomputational */
678 /* ===================================================================== */
679 /* Subroutine */ int dsytrf_aa_2stage_(char *uplo, integer *n, doublereal *a,
680 integer *lda, doublereal *tb, integer *ltb, integer *ipiv, integer *
681 ipiv2, doublereal *work, integer *lwork, integer *info)
683 /* System generated locals */
684 integer a_dim1, a_offset, i__1, i__2, i__3;
686 /* Local variables */
687 integer ldtb, i__, j, k;
688 extern /* Subroutine */ int dgemm_(char *, char *, integer *, integer *,
689 integer *, doublereal *, doublereal *, integer *, doublereal *,
690 integer *, doublereal *, doublereal *, integer *);
691 extern logical lsame_(char *, char *);
693 extern /* Subroutine */ int dcopy_(integer *, doublereal *, integer *,
694 doublereal *, integer *), dswap_(integer *, doublereal *, integer
695 *, doublereal *, integer *), dtrsm_(char *, char *, char *, char *
696 , integer *, integer *, doublereal *, doublereal *, integer *,
697 doublereal *, integer *);
700 integer i2, jb, kb, nb, td, nt;
701 extern /* Subroutine */ int dgbtrf_(integer *, integer *, integer *,
702 integer *, doublereal *, integer *, integer *, integer *),
703 dgetrf_(integer *, integer *, doublereal *, integer *, integer *,
704 integer *), dlacpy_(char *, integer *, integer *, doublereal *,
705 integer *, doublereal *, integer *), xerbla_(char *,
707 extern integer ilaenv_(integer *, char *, char *, integer *, integer *,
708 integer *, integer *, ftnlen, ftnlen);
709 extern /* Subroutine */ int dlaset_(char *, integer *, integer *,
710 doublereal *, doublereal *, doublereal *, integer *),
711 dsygst_(integer *, char *, integer *, doublereal *, integer *,
712 doublereal *, integer *, integer *);
713 logical tquery, wquery;
717 /* -- LAPACK computational routine (version 3.8.0) -- */
718 /* -- LAPACK is a software package provided by Univ. of Tennessee, -- */
719 /* -- Univ. of California Berkeley, Univ. of Colorado Denver and NAG Ltd..-- */
724 /* ===================================================================== */
727 /* Test the input parameters. */
729 /* Parameter adjustments */
731 a_offset = 1 + a_dim1 * 1;
740 upper = lsame_(uplo, "U");
741 wquery = *lwork == -1;
743 if (! upper && ! lsame_(uplo, "L")) {
747 } else if (*lda < f2cmax(1,*n)) {
749 } else if (*ltb < *n << 2 && ! tquery) {
751 } else if (*lwork < *n && ! wquery) {
757 xerbla_("DSYTRF_AA_2STAGE", &i__1, (ftnlen)16);
761 /* Answer the query */
763 nb = ilaenv_(&c__1, "DSYTRF_AA_2STAGE", uplo, n, &c_n1, &c_n1, &c_n1, (
764 ftnlen)16, (ftnlen)1);
767 tb[1] = (doublereal) ((nb * 3 + 1) * *n);
770 work[1] = (doublereal) (*n * nb);
773 if (tquery || wquery) {
783 /* Determine the number of the block size */
786 if (ldtb < nb * 3 + 1) {
789 if (*lwork < nb * *n) {
793 /* Determine the number of the block columns */
795 nt = (*n + nb - 1) / nb;
799 /* Initialize vectors/matrices */
802 for (j = 1; j <= i__1; ++j) {
808 tb[1] = (doublereal) nb;
812 /* ..................................................... */
813 /* Factorize A as U**T*D*U using the upper triangle of A */
814 /* ..................................................... */
817 for (j = 0; j <= i__1; ++j) {
819 /* Generate Jth column of W and H */
822 i__2 = nb, i__3 = *n - j * nb;
823 kb = f2cmin(i__2,i__3);
825 for (i__ = 1; i__ <= i__2; ++i__) {
827 /* H(I,J) = T(I,I)*U(I,J) + T(I,I+1)*U(I+1,J) */
834 dgemm_("NoTranspose", "NoTranspose", &nb, &kb, &jb, &
835 c_b12, &tb[td + 1 + i__ * nb * ldtb], &i__3, &a[(
836 i__ - 1) * nb + 1 + (j * nb + 1) * a_dim1], lda, &
837 c_b13, &work[i__ * nb + 1], n);
839 /* H(I,J) = T(I,I-1)*U(I-1,J) + T(I,I)*U(I,J) + T(I,I+1)*U(I+1,J) */
846 dgemm_("NoTranspose", "NoTranspose", &nb, &kb, &jb, &
847 c_b12, &tb[td + nb + 1 + (i__ - 1) * nb * ldtb], &
848 i__3, &a[(i__ - 2) * nb + 1 + (j * nb + 1) *
849 a_dim1], lda, &c_b13, &work[i__ * nb + 1], n);
856 dlacpy_("Upper", &kb, &kb, &a[j * nb + 1 + (j * nb + 1) * a_dim1],
857 lda, &tb[td + 1 + j * nb * ldtb], &i__2);
859 /* T(J,J) = U(1:J,J)'*H(1:J) */
862 dgemm_("Transpose", "NoTranspose", &kb, &kb, &i__2, &c_b21, &
863 a[(j * nb + 1) * a_dim1 + 1], lda, &work[nb + 1], n, &
864 c_b12, &tb[td + 1 + j * nb * ldtb], &i__3);
865 /* T(J,J) += U(J,J)'*T(J,J-1)*U(J-1,J) */
867 dgemm_("Transpose", "NoTranspose", &kb, &nb, &kb, &c_b12, &a[(
868 j - 1) * nb + 1 + (j * nb + 1) * a_dim1], lda, &tb[td
869 + nb + 1 + (j - 1) * nb * ldtb], &i__2, &c_b13, &work[
872 dgemm_("NoTranspose", "NoTranspose", &kb, &kb, &nb, &c_b21, &
873 work[1], n, &a[(j - 2) * nb + 1 + (j * nb + 1) *
874 a_dim1], lda, &c_b12, &tb[td + 1 + j * nb * ldtb], &
879 dsygst_(&c__1, "Upper", &kb, &tb[td + 1 + j * nb * ldtb], &
880 i__2, &a[(j - 1) * nb + 1 + (j * nb + 1) * a_dim1],
884 /* Expand T(J,J) into full format */
887 for (i__ = 1; i__ <= i__2; ++i__) {
889 for (k = i__ + 1; k <= i__3; ++k) {
890 tb[td + (k - i__) + 1 + (j * nb + i__ - 1) * ldtb] = tb[
891 td - (k - (i__ + 1)) + (j * nb + k - 1) * ldtb];
902 dgemm_("NoTranspose", "NoTranspose", &kb, &kb, &kb, &
903 c_b12, &tb[td + 1 + j * nb * ldtb], &i__2, &a[
904 (j - 1) * nb + 1 + (j * nb + 1) * a_dim1],
905 lda, &c_b13, &work[j * nb + 1], n);
909 dgemm_("NoTranspose", "NoTranspose", &kb, &kb, &i__2,
910 &c_b12, &tb[td + nb + 1 + (j - 1) * nb * ldtb]
911 , &i__3, &a[(j - 2) * nb + 1 + (j * nb + 1) *
912 a_dim1], lda, &c_b13, &work[j * nb + 1], n);
915 /* Update with the previous column */
917 i__2 = *n - (j + 1) * nb;
919 dgemm_("Transpose", "NoTranspose", &nb, &i__2, &i__3, &
920 c_b21, &work[nb + 1], n, &a[((j + 1) * nb + 1) *
921 a_dim1 + 1], lda, &c_b12, &a[j * nb + 1 + ((j + 1)
922 * nb + 1) * a_dim1], lda);
925 /* Copy panel to workspace to call DGETRF */
928 for (k = 1; k <= i__2; ++k) {
929 i__3 = *n - (j + 1) * nb;
930 dcopy_(&i__3, &a[j * nb + k + ((j + 1) * nb + 1) * a_dim1]
931 , lda, &work[(k - 1) * *n + 1], &c__1);
934 /* Factorize panel */
936 i__2 = *n - (j + 1) * nb;
937 dgetrf_(&i__2, &nb, &work[1], n, &ipiv[(j + 1) * nb + 1], &
939 /* IF (IINFO.NE.0 .AND. INFO.EQ.0) THEN */
940 /* INFO = IINFO+(J+1)*NB */
943 /* Copy panel back */
946 for (k = 1; k <= i__2; ++k) {
947 i__3 = *n - (j + 1) * nb;
948 dcopy_(&i__3, &work[(k - 1) * *n + 1], &c__1, &a[j * nb +
949 k + ((j + 1) * nb + 1) * a_dim1], lda);
952 /* Compute T(J+1, J), zero out for GEMM update */
955 i__2 = nb, i__3 = *n - (j + 1) * nb;
956 kb = f2cmin(i__2,i__3);
958 dlaset_("Full", &kb, &nb, &c_b13, &c_b13, &tb[td + nb + 1 + j
959 * nb * ldtb], &i__2);
961 dlacpy_("Upper", &kb, &nb, &work[1], n, &tb[td + nb + 1 + j *
965 dtrsm_("R", "U", "N", "U", &kb, &nb, &c_b12, &a[(j - 1) *
966 nb + 1 + (j * nb + 1) * a_dim1], lda, &tb[td + nb
967 + 1 + j * nb * ldtb], &i__2);
970 /* Copy T(J,J+1) into T(J+1, J), both upper/lower for GEMM */
974 for (k = 1; k <= i__2; ++k) {
976 for (i__ = 1; i__ <= i__3; ++i__) {
977 tb[td - nb + k - i__ + 1 + (j * nb + nb + i__ - 1) *
978 ldtb] = tb[td + nb + i__ - k + 1 + (j * nb +
982 dlaset_("Lower", &kb, &nb, &c_b13, &c_b12, &a[j * nb + 1 + ((
983 j + 1) * nb + 1) * a_dim1], lda);
985 /* Apply pivots to trailing submatrix of A */
988 for (k = 1; k <= i__2; ++k) {
990 ipiv[(j + 1) * nb + k] += (j + 1) * nb;
992 i1 = (j + 1) * nb + k;
993 i2 = ipiv[(j + 1) * nb + k];
995 /* > Apply pivots to previous columns of L */
997 dswap_(&i__3, &a[(j + 1) * nb + 1 + i1 * a_dim1], &
998 c__1, &a[(j + 1) * nb + 1 + i2 * a_dim1], &
1000 /* > Swap A(I1+1:M, I1) with A(I2, I1+1:M) */
1003 dswap_(&i__3, &a[i1 + (i1 + 1) * a_dim1], lda, &a[
1004 i1 + 1 + i2 * a_dim1], &c__1);
1006 /* > Swap A(I2+1:M, I1) with A(I2+1:M, I2) */
1009 dswap_(&i__3, &a[i1 + (i2 + 1) * a_dim1], lda, &a[
1010 i2 + (i2 + 1) * a_dim1], lda);
1012 /* > Swap A(I1, I1) with A(I2, I2) */
1013 piv = a[i1 + i1 * a_dim1];
1014 a[i1 + i1 * a_dim1] = a[i2 + i2 * a_dim1];
1015 a[i2 + i2 * a_dim1] = piv;
1016 /* > Apply pivots to previous columns of L */
1019 dswap_(&i__3, &a[i1 * a_dim1 + 1], &c__1, &a[i2 *
1020 a_dim1 + 1], &c__1);
1028 /* ..................................................... */
1029 /* Factorize A as L*D*L**T using the lower triangle of A */
1030 /* ..................................................... */
1033 for (j = 0; j <= i__1; ++j) {
1035 /* Generate Jth column of W and H */
1038 i__2 = nb, i__3 = *n - j * nb;
1039 kb = f2cmin(i__2,i__3);
1041 for (i__ = 1; i__ <= i__2; ++i__) {
1043 /* H(I,J) = T(I,I)*L(J,I)' + T(I+1,I)'*L(J,I+1)' */
1050 dgemm_("NoTranspose", "Transpose", &nb, &kb, &jb, &c_b12,
1051 &tb[td + 1 + i__ * nb * ldtb], &i__3, &a[j * nb +
1052 1 + ((i__ - 1) * nb + 1) * a_dim1], lda, &c_b13, &
1053 work[i__ * nb + 1], n);
1055 /* H(I,J) = T(I,I-1)*L(J,I-1)' + T(I,I)*L(J,I)' + T(I,I+1)*L(J,I+1)' */
1057 jb = (nb << 1) + kb;
1062 dgemm_("NoTranspose", "Transpose", &nb, &kb, &jb, &c_b12,
1063 &tb[td + nb + 1 + (i__ - 1) * nb * ldtb], &i__3, &
1064 a[j * nb + 1 + ((i__ - 2) * nb + 1) * a_dim1],
1065 lda, &c_b13, &work[i__ * nb + 1], n);
1069 /* Compute T(J,J) */
1072 dlacpy_("Lower", &kb, &kb, &a[j * nb + 1 + (j * nb + 1) * a_dim1],
1073 lda, &tb[td + 1 + j * nb * ldtb], &i__2);
1075 /* T(J,J) = L(J,1:J)*H(1:J) */
1076 i__2 = (j - 1) * nb;
1078 dgemm_("NoTranspose", "NoTranspose", &kb, &kb, &i__2, &c_b21,
1079 &a[j * nb + 1 + a_dim1], lda, &work[nb + 1], n, &
1080 c_b12, &tb[td + 1 + j * nb * ldtb], &i__3);
1081 /* T(J,J) += L(J,J)*T(J,J-1)*L(J,J-1)' */
1083 dgemm_("NoTranspose", "NoTranspose", &kb, &nb, &kb, &c_b12, &
1084 a[j * nb + 1 + ((j - 1) * nb + 1) * a_dim1], lda, &tb[
1085 td + nb + 1 + (j - 1) * nb * ldtb], &i__2, &c_b13, &
1088 dgemm_("NoTranspose", "Transpose", &kb, &kb, &nb, &c_b21, &
1089 work[1], n, &a[j * nb + 1 + ((j - 2) * nb + 1) *
1090 a_dim1], lda, &c_b12, &tb[td + 1 + j * nb * ldtb], &
1095 dsygst_(&c__1, "Lower", &kb, &tb[td + 1 + j * nb * ldtb], &
1096 i__2, &a[j * nb + 1 + ((j - 1) * nb + 1) * a_dim1],
1100 /* Expand T(J,J) into full format */
1103 for (i__ = 1; i__ <= i__2; ++i__) {
1105 for (k = i__ + 1; k <= i__3; ++k) {
1106 tb[td - (k - (i__ + 1)) + (j * nb + k - 1) * ldtb] = tb[
1107 td + (k - i__) + 1 + (j * nb + i__ - 1) * ldtb];
1114 /* Compute H(J,J) */
1118 dgemm_("NoTranspose", "Transpose", &kb, &kb, &kb, &
1119 c_b12, &tb[td + 1 + j * nb * ldtb], &i__2, &a[
1120 j * nb + 1 + ((j - 1) * nb + 1) * a_dim1],
1121 lda, &c_b13, &work[j * nb + 1], n);
1125 dgemm_("NoTranspose", "Transpose", &kb, &kb, &i__2, &
1126 c_b12, &tb[td + nb + 1 + (j - 1) * nb * ldtb],
1127 &i__3, &a[j * nb + 1 + ((j - 2) * nb + 1) *
1128 a_dim1], lda, &c_b13, &work[j * nb + 1], n);
1131 /* Update with the previous column */
1133 i__2 = *n - (j + 1) * nb;
1135 dgemm_("NoTranspose", "NoTranspose", &i__2, &nb, &i__3, &
1136 c_b21, &a[(j + 1) * nb + 1 + a_dim1], lda, &work[
1137 nb + 1], n, &c_b12, &a[(j + 1) * nb + 1 + (j * nb
1138 + 1) * a_dim1], lda);
1141 /* Factorize panel */
1143 i__2 = *n - (j + 1) * nb;
1144 dgetrf_(&i__2, &nb, &a[(j + 1) * nb + 1 + (j * nb + 1) *
1145 a_dim1], lda, &ipiv[(j + 1) * nb + 1], &iinfo);
1146 /* IF (IINFO.NE.0 .AND. INFO.EQ.0) THEN */
1147 /* INFO = IINFO+(J+1)*NB */
1150 /* Compute T(J+1, J), zero out for GEMM update */
1153 i__2 = nb, i__3 = *n - (j + 1) * nb;
1154 kb = f2cmin(i__2,i__3);
1156 dlaset_("Full", &kb, &nb, &c_b13, &c_b13, &tb[td + nb + 1 + j
1157 * nb * ldtb], &i__2);
1159 dlacpy_("Upper", &kb, &nb, &a[(j + 1) * nb + 1 + (j * nb + 1)
1160 * a_dim1], lda, &tb[td + nb + 1 + j * nb * ldtb], &
1164 dtrsm_("R", "L", "T", "U", &kb, &nb, &c_b12, &a[j * nb +
1165 1 + ((j - 1) * nb + 1) * a_dim1], lda, &tb[td +
1166 nb + 1 + j * nb * ldtb], &i__2);
1169 /* Copy T(J+1,J) into T(J, J+1), both upper/lower for GEMM */
1173 for (k = 1; k <= i__2; ++k) {
1175 for (i__ = 1; i__ <= i__3; ++i__) {
1176 tb[td - nb + k - i__ + 1 + (j * nb + nb + i__ - 1) *
1177 ldtb] = tb[td + nb + i__ - k + 1 + (j * nb +
1181 dlaset_("Upper", &kb, &nb, &c_b13, &c_b12, &a[(j + 1) * nb +
1182 1 + (j * nb + 1) * a_dim1], lda);
1184 /* Apply pivots to trailing submatrix of A */
1187 for (k = 1; k <= i__2; ++k) {
1189 ipiv[(j + 1) * nb + k] += (j + 1) * nb;
1191 i1 = (j + 1) * nb + k;
1192 i2 = ipiv[(j + 1) * nb + k];
1194 /* > Apply pivots to previous columns of L */
1196 dswap_(&i__3, &a[i1 + ((j + 1) * nb + 1) * a_dim1],
1197 lda, &a[i2 + ((j + 1) * nb + 1) * a_dim1],
1199 /* > Swap A(I1+1:M, I1) with A(I2, I1+1:M) */
1202 dswap_(&i__3, &a[i1 + 1 + i1 * a_dim1], &c__1, &a[
1203 i2 + (i1 + 1) * a_dim1], lda);
1205 /* > Swap A(I2+1:M, I1) with A(I2+1:M, I2) */
1208 dswap_(&i__3, &a[i2 + 1 + i1 * a_dim1], &c__1, &a[
1209 i2 + 1 + i2 * a_dim1], &c__1);
1211 /* > Swap A(I1, I1) with A(I2, I2) */
1212 piv = a[i1 + i1 * a_dim1];
1213 a[i1 + i1 * a_dim1] = a[i2 + i2 * a_dim1];
1214 a[i2 + i2 * a_dim1] = piv;
1215 /* > Apply pivots to previous columns of L */
1218 dswap_(&i__3, &a[i1 + a_dim1], lda, &a[i2 +
1224 /* Apply pivots to previous columns of L */
1226 /* CALL DLASWP( J*NB, A( 1, 1 ), LDA, */
1227 /* $ (J+1)*NB+1, (J+1)*NB+KB, IPIV, 1 ) */
1232 /* Factor the band matrix */
1233 dgbtrf_(n, n, &nb, &nb, &tb[1], &ldtb, &ipiv2[1], info);
1237 /* End of DSYTRF_AA_2STAGE */
1239 } /* dsytrf_aa_2stage__ */